Radeon HD 5870 & 5850 PPD projections

By , the in Folding@Home Project - No comment
Radeon HD 5870 & 5850 PPD projections
Initial user reports regarding client performance and PPD output on the GPU2 client are beginning to emerge.

It seems that in its default configuration, the HD 5870 produces 4600 PPD on a WU with 973 atoms (384 points). After setting certain environmental variables, it is possible for the same card to produce 6000 PPD.

In neither cases was information obtained on the operating system being run, or the environmental variables which were set. For comparison, my HD 4870 and its 800MHz GPU clock (with properly configured environmental variables) gets 2700 PPD on a unit of 973 atoms, and 3400 PPD on a unit of 1392 atoms.

All hopes are now on the GPU3 client, which we remind you uses OpenCL instead of Stream/CAL. We'll keep you posted when we have other performance benchmarks, and more information on the GPU3 client.

BigAdv projects are making progress

By , the in Folding@Home Project - No comment
In this post on the official Folding@Home support forums, Peter Kasson tells us that BigAdv projects are doing fine and produce interesting results.

More specifically, thanks to the results produced by first sets of projects, some scientific additions are being worked on to get the core ready fo the next set of projects in this category. This is the normal process of constant improvement of the folding cores. We will probably see these improvements in the next core update.

This is very encouraging and it suggests that BigAdv projects will continue in the near future.

New nVidia Core 11 v1.31 in testing

By , the in Folding@Home Project - 1 Comment
Mark Friedrichs has posted a new version of the nVidia GPU2 core 11, aiming to fix the problems people have been having running multi-GPU systems with mismatched cards (mismatched shader count or boards from different generations). It also contains updates to the checkpointing system to prevent previous checkpoints left from old units from corrupting the current work unit.

This core contains all previous fixes and can be downloaded from here if you wish to dine in hell test it. It is version 1.31. Please only upgrade the core BETWEEN units as it is likely that the changes to the checkpointing logic will destroy all progress on the current unit if the switch is made mid-unit.

Update on New Cores and Clients

By , the in Folding@Home Project - No comment
Stanford has been very busy with new cores and clients for FAH, and Vijay Pande has published some new details.

The new announcements concern:
  • The two cores for SMP2
  • The Protomol core and a new MD method, NML
  • GPU3: OpenMM and OpenCL

The full story can be found here.

Breaking news: HD5800 Performance and OpenCL's future

By , the in Folding@Home Project - No comment
Breaking news: HD5800 Performance and OpenCL's future
mhouston, AKA Michael Houston - the developer of the ATI GPU2 core - has just spoken about the future folding performance of the Radeon 5800. He has already tested the card, and the expected benefits are as follows:

  • Up to 50% faster performance on large WUs (not distributed as of yet)
  • A small performance increase on current small WUs

Obviously, these figures are only realised with certain optimisations enabled, since the core is optimised for the Radeon HD4800 and 800 vector processors. However, it seems that Michael no longer wishes to change the current core.

The current core - based on CAL / Brook+ / Stream - has become impossible to maintain. The CAL (Compute Abstraction Layer), which is a layer of abstraction between the application and hardware, is not his work; when new hardware comes out, the core is further optimised for it. For Michael, the future of folding is on ATI's OpenCL, because the layer is much more mature. He wants to take over the development of a new core built on this new API, to avoid having the problems that have plagued the current core.

The direct consequence is that nVidia and Intel could also be supported in the future.

We aren't able to say whether work has yet begun, but we follow new developments with anticipation.

Translated by : KaySL

Source: Folding Forum

A2 SMP projects fixed.

By , the in Folding@Home Project - No comment
The repairing and rebuilding process of the SMP A2 WUs affected by folding on 1 core bug is now completed.

Simulations of the affected trajectories should now be possible and the number of “CoreStatus FF (255)” errors (the sign of this bug since 2.10 core) that you might encounter would be drastically reduced.

About core 2.10 performances on dual core processors, you’ll have to get used to them. :( We hope that performance will be back to the previous level in future core updates. My advice, if you can, is to replace your dual core VMs with native Linux or quad core VMs, waiting for better days.

Cleaning up bad SMP A2 WUs.

By , the in Folding@Home Project - No comment
The news is quite good : Peter Kasson is currently running a script server side that is repairing all trajectories affected by the "fold on 1 core" issue (that produces CoreStatus FF errors with 2.10 A2 core and has a ~1.5 MB data file).

To make it simple, the script is removing affected WUs ( "Gen N" WUs) and previous generation WUs ("Gen N-1" WUs) that produced these erroneous WUs. With this script, we should be able to refold those Gen N-1 WUs and then Gen N WUs and thanks to core 2.10 fixes, they should fold fine and keep the trajectories running.

Project 2677 has been cleared and Peter Kasson has a script currently working on Project 2671. Unfortunately, this process is not entirely automated and it doesn't automatically regenerate new WUs. This might cause a temporary shortage of A2 WUs until the rebuild process is completed.

Don't be surprised if you fold some A1 WUs or if you have trouble getting work while the rebuild process is running.

New Mac OS X client posted

By , the in Folding@Home Project - No comment
New Mac OS X client posted
If you are a Mac OS X user, a new Folding@home client has been released for you.

The news for the two previous versions (6.26 and 6.26.1) can be found here.

The new features for v6.26.2 are as follows:
  • The client should now run on both 10.5 Leopard and 10.6 Snow Leopard.
  • Further corrections have been made to the launchd script and installer.
  • Further corrections to the prefpane (courtesy of Calxalot).
  • The system preferences panel should no longer need to restart in 32-bit mode. (for 10.6)

This client is the version with installer that integrates into your OS System preferences, and starts automatically with your OS. To download version 6.26.2, click here.

How many atoms are there in a BigAdv unit?

By , the in Folding@Home Project - No comment
Some of you may have noticed that the Project Summary page is failing to display the atom count for BigAdv units (p268x). If you are wondering how many atoms these units contain then you need wonder no more...

toTOW decided to investigate after reviewing the introduction of Peter Kasson, especially this passage:

Indeed, due to the very large size of lipid vesicle simulations (which can easily be 1 to 10 million atoms in size), the SMP client is extremely important to this research.

He finally found the answer after reviewing a set of log files, which contain details for the researchers (these files are sent back in the results package) which are not printed in FAHlog.txt... the BigAdv units contain at least 1,094,681 atoms!

More information on these projects: they study the process of cell infection by the influenza virus (see description), which involves the fusion of liposomes (the fatty covering of the virus fusing with the fatty covering of the cell being infected). This process is illustrated in the following image:

Liposome fusion

With over a million atoms in the simulation, one can easily understand the need for massive amounts of memory and computing power when processing BigAdv units. The conventional SMP units are comparative featherweights, with a "mere" 147,087 atoms!

A focus on the SMP A2 2.10/2.11 core

By , the in Folding@Home Project - No comment
It's been several days since the 2.10 core was released, to fix the bug which was creating defective folding units on single cores. It is time to take stock of all that has changed with this core.

Note also that the A2 core is now at version 2.11 for the Mac; debug output had been forgotten and not been removed, which produced a great deal of extraneous data to be written to the logs. This error has been corrected in the latest version.

The list of changes in this version is quite long; click on "Read More" to view the complete article.

Translated by : KaySL